International Journal of Trend in Scientific
Research and Development (IJTSRD)
International Open Access Journal
ISSN No: 2456 - 6470 | www.ijtsrd.com | Volume - 2 | Issue – 1
Diagnosis of hyperglycemia u
using
sing
Artificial Neural Networks
Abid Sarwar
Department of Computer S
Science & IT (Bhaderwah Campus)
University of Jammu, Jammu, India
ABSTRACT
The aim of Artificial Intelligence is to develop the
machines to perform the tasks in a better way than the
humans. Another aim of Artificial Intelligence is to
understand the actions whether it occurs in humans,
machines or animals. As a result, Artificial
Intelligence is gaining Importance in science and
engineering fields. The use of Artificial Intelligence in
medicall diagnosis too is becoming increasingly
common and has been used widely in the diagnosis of
cancers, tumors, hepatitis, lung diseases, etc... The
main aim of this paper is to build an Artificial
Intelligent System that after analysis of certain
parameters can predict that whether a person is
diabetic or not. Diabetes is the name used to describe
a metabolic condition of having higher than normal
blood sugar levels. Diabetes is becoming increasingly
more common throughout the world, due to increased
obesity - which can lead to metabolic syndrome or
pre-diabetes
diabetes leading to higher incidences of type 2
diabetes. Authors have identified 10 parameters that
play an important role in diabetes and prepared a rich
database of training data which served as the
backbonee of the prediction algorithm. Keeping in
view this training data authors developed a system
that uses the artificial neural networks algorithm to
serve the purpose. These are capable of predicting
new observations (on specific variables) from
previous observations
ervations (on the same or other variables)
after executing a process of so-called
called learning from
existing training data (Haykin1998).The results
indicate that the ANN is the best predictor with the
accuracy of about 96%. This system can be used to
assist medical
dical programs especially in geographically
remote areas where expert human diagnosis not
possible with an advantage of minimal expenses and
faster results.
Keywords: Artificial Intelligence, metabolic, Machine
Learning, Diabetes, Artificial Neural Network,
Medical Diagnosis.
I.
INTRODUCTION
Diabetes is a disease that occurs when blood glucose,
also called blood sugar, is too high. Blood glucose is
main source of energy and comes from the food we
eat. Insulin, a hormone made by the pancreas, helps
glucose
ose from food get into your cells to be used for
energy. Sometime body doesn’t make enough—or
enough
any—insulin
insulin or doesn’t use insulin well. Glucose then
stays in blood and doesn’t reach cells. Diabetes can
mainly be of 3 types: Type-1
Type
diabetes, Type-2
diabetes and Gestational diabetes. Type-1
Type diabetes
results from non-production
production of insulin & Type-2
Type
diabetes results from development of resistance of
insulin, as a result of which the insulin produced is
not able to metabolize the sugar levels properly.
Gestational diabetes occurs in pregnant women, who
develop a high blood glucose level during pregnancy
who never had any previous such history. It may be
preceded by development of type-2
type
diabetes. As
reported by WHO & International Diabetic Federation
in year 2010, the toll of diabetic patients was 285
million and this number is expected to grow to 483
million by 2030. WHO estimates that between 2010
to 2030 there will be an increase of 69% in adult
diabetic population in developing countries and 20%
increase of the same in developed countries. India
@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 2 | Issue – 1 | Nov-Dec
Dec 2017
Page: 606
International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
having 50.8 million patients of this disease leads the
world & is followed by China (43.2), United States
(26.8). In 2010 diabetes caused 3.9 million deaths
worldwide. The primary concern of AI in medicine is
the construction of AI programs that can assert a
medical doctor in performing expert diagnosis. These
programs by making use of various computational
sciences such as statistics and probability find out the
hidden patterns from the training data and using these
patterns they classify the test data into one the
possible categories. The backbones of these AI
programs are the various data sets prepared from
various clinical cases which act as practical examples
in training the system. The decision and
recommendation prepared from these systems can be
illustrated to the subjects after combining them with
the experience of human expert.
recognition, computer vision, and also in medical
diagnosis. The ability of ANNs to recognize and
classify various patterns accurately has attracted the
attention of many researchers to solve problems from
clinical domain. In last 20 years, ANN has been most
popularly used AI technique in medical data mining
(Ramesh et al. 2004). ANNs are typically
characterized by three parameters: (1) Pattern of
interconnection between different artificial neurons.
The neurons are grouped into different layers,
typically an input layer, one or more hidden layers,
and an output layer. (2) The learning, which is
achieved by repeatedly adjusting the numerical
weights associated with the interconnecting edges
between different artificial neurons. (3) Activation
function that converts a neuron’s weighted input to its
output activation.
2. METHODOLOGY
3. PREVIOUS WORK
An artificial neural network (ANN) is a mathematical
model which is inspired by the biological network of
neurons. An ANN is a powerful data-modeling tool
that is able to capture and represent and simulate
complex relationships between inputs and outputs by
performing multiple parallel computations. ANNs are
computational analytic tools, modeled after the
processes of ‘‘learning’’ in the cognitive system and
the neurological functions of the brain. These are
capable of predicting new observations (on specific
variables) from previous observations (on the same or
other variables) after executing a process of so-called
learning from existing training data (Haykin 1998).
ANN consists of a group of highly interconnected
processing elements called ‘‘artificial neurons’’ and it
processes information using a connectionist approach
to computation. An ANN consists of a pool of
processing units which communicate with each other
by sending signals over a large number of weighted
connections. ANN is an adaptive system that changes
and adapts its structure during a learning phase.
ANNs have been shown to perform more accurately
as compared to its counterparts because of their ability
to deal with incomplete and noisy data (Mahamud et
al. 1999). One of the key elements of a neural network
is its ability to learn. A neural network is complex
adaptive system, that is, it can change its internal
structure based on the information flowing through it.
The predictive ability of an ANN can be improved by
iterative learning algorithms. ANNs are now days
widely being used in solving problems of type, ‘‘easyfor-a-human, difficult-for-a-machine’’ such as various
problems in domain of pattern recognition, speech
It has been noted that the machine learning algorithms
are increasingly being used in solving problems in
Medical Domains such as in Oncology (Lisboa and
Taktak 2006; Catto et al. 2003; Anagnostou et al.
2003; Bratko and Kononenko 1987), Urology
(Anagnostou et al. 2003), Liver Pathology (Lesmo et
al. 1983), Cardiology (Catlett 1991; Clark and
Boswell 1991), Gynecology (Nunez 1990), Thyroid
disorders (Hojker et al. 1998; Horn et al. 1985), and
Perinatology (Kern et al. 1990). Numerous
researchers have used AI and data mining-based
algorithms to solve problems in the field of medicine.
Chiu et al. (2008) studied the application of artificial
neural networks in predicting the skeletal metastasis
in patients suffering from prostate cancer. They
analyzed whole body bone scintigraphies of patients
with prostate cancer who underwent the technetium99 m methylene diphosphate (tc-99 m MDP) for a
period of five years. For the analysis, they used an
artificial neural network having four layers of
perceptrons which was trained with the data set of 111
prostate cancer patients. For evaluating the
performance
of
ANN,
receiver
operating
characteristics (ROC) analysis was used. The area
under the ROC curve (0.88 ± 0.07) revealed excellent
discriminatory power (p\0.001) with the best
simultaneous sensitivity (87.5 %) and specificity (83.3
%). Authors concluded that an ANN, which is based
on limited clinical parameters, to be a promising
method in forecasting of the skeletal metastasis.
@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 2 | Issue – 1 | Nov-Dec 2017
Page: 607
International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
4. PARAMETERS USED IN ESTIMATION
the sake of maintaining consistency. Table-I
summaries the parameters chosen and their allowed
values. A dataset of 415 cases was prepared by
collecting the data randomly from different sections
of the society with an aim to have a variety in the
dataset. To maintain accuracy and to avoid errors,
considerable care was taken to ensure that the
database had correct values.
Since India is having the highest Diabetic population
in the world so it was easy to collect the data about
the patients who suffered from this disease. After a
detailed study, authors identified ten best
physiological parameters for the study which were so
chosen that the values for them could be easily
determined and could be assigned discrete values, for
Table-I: Various parameters used and their allowed values
Parameter
Age
Description
Age of the subject
Family History
Whether any family member of the subject is suffering/
was suffering from diabetes.
Whether male or female
Whether the subject does smoking or not.
Whether the subject does drinking or not.
Does a person feel tired after doing a little work?
Whether the subject frequently feels a strong desire to
drink water. i.e how many times the subject drinks water.
How many times the subject passes urine in a day
Sex
Smoking
Drinking
Fatigue
Thurst
Frequency of
urination
Height
Weight
Height of the subject
Weight of the subject
Allowed Values
Discrete Integer Values
Yes or No
Male or Female
Yes or No
Yes or No
Yes or No
Discrete Integer values
Discrete Integer values
Discrete floating point
values
Discrete floating point
values
5. IMPLEMENTATION
In the present study, we have used the standard
backpropagation-based multilayer perceptron (MLP)
architecture of ANN. This architecture is most
commonly used architecture for ANN in medical
research. It consists of multiple layers of artificial
neurons arranged in a directed graph, having each
layer being fully connected to the next layer. All these
are connected in a feed forward fashion with input
layer connected to hidden layer and hidden layer
connected to output layer. Backpropagation is a
gradient descent technique to minimize the error
criteria and it is simply a way to determine the error
values in hidden layers. A hidden layer allows ANN
to develop its representation of input–output mapping.
In backpropagation, error data at the output layer are
‘‘back propagated’’ to earlier ones, allowing
incoming weights to these layers to be updated and
adjusted such that the error between the input and
desired output is as least as possible. For processing
the inputs at the artificial neurons, Sigmoid function
was used as the activation function. Forrealization of
desired ANN, we used the Neural Network Toolbox
of Matlab 7.6.0. Figure 1(a) shows the topology for
the neural network used in the study and Figure 1(b)
shows the Matlab program in execution. Numbers of
artificial neurons in the input, hidden, and output
layers are 10, 20, and 1, respectively. To build the
model, we repeated the training process up to one
million times or under 0.0001 for the mean squared
error in the model. The sigmoid function was
employed, and the weight was decided randomly in
the connections for the network.
@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 2 | Issue – 1 | Nov-Dec 2017
Page: 608
International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
2456
6. CONCLUSION
This artificial neural network based system is very
useful for diagnosis of diabetes. The reliability of the
system was evaluated by predicting new observations
(on specific variables) from previous observations (on
the same or other variables) after executing a process
of so-called
called learning from existing training data. The
results suggest that this system can perform good
prediction with least error and finally this techniqu
technique
could be an important tool for supplementing the
medical doctors in performing expert diagnosis. In
this method the efficiency of Forecasting was found to
be around 96%. Its performance can be further
improved by identifying & incorporating various
other parameters and increasing the size of training
data.
REFERENCE
1) Anagnostou T, Remzi M, Lykourinas M, Djavan
B (2003) Artificial neural networks for decision
decisionmaking in urologiconcology. EurUrol 43(6):596
43(6):596–
603
2) Bharat Rao R, Bi J, Obuchowski N, Naidich D
(2007)
007) LungCAD: a clinically approved, machine
learning system for lung cancer detection. In:
International conference on knowledge discovery
and data mining 2007, San Jose, California, USA,
ACM 978-1-59593-609-7/07/0008
7/07/0008
3) Bonham GS, Brock DB (1985) The relationship
relati
of
diabetes with race, sex, and obesity. Am J
ClinNutr 41:776–783
4) Bratko I, Kononenko I (1987) Learning rules from
incomplete and noisy data. In: Phelip B (ed)
Interactions in artificial intelligence and statistical
methods. Technical Press, Hampshire
Hampshi
5) Caruana R, Niculescu--Mizil A (2006) An
empirical comparison of supervised learning
algorithms. In: ICML ‘06 Proceedings of the 23rd
international conference on machine learning, pp
161–168
@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 2 | Issue – 1 | Nov-Dec
Dec 2017
Page: 609
International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
6) Catlett J (1991) On changing continuous attributes
into ordered discrete attributes. In: Proceedings of
European working session on learning-91,
Portugal, 4–6 Mar 1991, pp 164–178
9) Chang C-L,HsuM-Y(2009) The study that applies
artificial intelligence and logistic regression for
assistance in differential diagnostic of pancreatic
cancer. Expert SystAppl 36(7):10663–10672
7) Catto JWF, Linkens DA, Abbod MF, Chen M,
Burton JL, Feeley KM, Hamdy FC (2003)
Artificial intelligence in predicting bladder cancer
outcome: a comparison of neuro-fuzzy modeling
and artificial neural networks. Clin Cancer Res
9:4172
10) Chiu J-S, Wang Y-F, Su Y-C,Wei L-H, Liao J-G,
LiY-C (2008)Artificial neural network to predict
skeletal metastasis in patients with prostate
cancer. Springer science ?Business Media, May
2008
8) Chan JM, Rimm EB, Colditz GA, Stampfer MJ,
Willett WC (1994) Obesity, fat distribution, and
weight gain as risk factors for clinical diabetes in
men. Diabetes Care 17(9):961–969
11) 11. Clark P, Boswell R (1991) Rule Induction
with CN2: some recent improvements. In:
Proceedings of European working session on
learning-91, Portugal, Mar 1991, pp 151–163
@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 2 | Issue – 1 | Nov-Dec 2017
Page: 610